Download Report

Lp NORMS, NODAL SETS, AND QUANTUM ERGODICITY
`
HAMID HEZARI AND GABRIEL RIVIERE
Abstract. For small range of p > 2, we improve the Lp bounds of eigenfunctions
of the Laplacian on negatively curved manifolds. Our improvement is by a power
of logarithm for a full density sequence of eigenfunctions. We also derive improvements on the size of the nodal sets. Our proof is based on a quantum ergodicity
property of independent interest, which holds for families of symbols supported in
balls whose radius shrinks at a logarithmic rate.
1. Introduction
Let (M, g) be a smooth boundaryless compact connected Riemannian manifold of
dimension d, and let ∆g be the Laplace-Beltrami operator. The eigenfunctions ψλ
of ∆g are nonzero solutions to
−∆g ψλ = λψλ ,
λ ≥ 0.
A classical problem is to study the asymptotic properties of ψλ as λ → +∞, and
their relations to the geometry of the manifold (M, g). For instance, one can study
the size of their Lp norms when ψλ is L2 normalized, or study the geometry of their
nodal sets
Zψλ = {x ∈ M ; ψλ (x) = 0}.
Another interesting problem is to study the asymptotic of the positive measures
|ψλ |2 dvg , where vg is the normalized volume measure on M . The purpose of this
article is to establish a new asymptotic property of these measures on negatively
curved manifolds, and to deduce from this some results on the Lp norms and the
size of nodal sets of eigenfunctions. This new property is that quantum ergodicity
1
holds on small scale balls of radius (log λ)−K with 0 < K < 2d
. One of the main
observations of the present article is that any improvement on the radius of such
balls would give improvements on Lp estimates, and on the size of nodal sets.
Before we state our results we recall that since M is compact, there exist isolated
eigenvalues with finite multiplicities
0 = λ1 < λ2 ≤ λ3 · · · → ∞,
Date: March 29, 2015.
1
2
`
HAMID HEZARI AND GABRIEL RIVIERE
and an orthonormal basis (ψj )j∈N of L2 (M ) satisfying
−∆g ψj = λj ψj .
1.1. Lp norms. Our first result gives upper bounds on the Lp norms of eigenfunctions on negatively curved manifolds. More precisely, we will show that Lp bounds
of eigenfunctions can be improved along a full density subsequence for small values
of p.
Our main result on Lp norms of eigenfunctions is the following:
Theorem 1.1. Let (M, g) be a compact negatively curved manifold. Then, for
every K ∈ (0, d1 ), there exists CK,p > 0 such that, for every ONB {ψj } of ∆g
eigenfunctions, there exists a full density subset S of N such that
�
�γ(p)
λj
(1)
j ∈ S, p ∈ (2, ∞] : �ψj �Lp (M ) ≤ CK,p
,
(log λj )K
where
�
�
�
�
d−1 1 1
d−1
d
γ(p) = max
−
,
−
,
4
2 p
4
2p
is the Sogge exponent.
Recall that a subset S of N is said to have full density if
card (S ∩ [1, N ])
lim
= 1.
N →∞
N
On a general smooth compact Riemannian manifold (without boundary), the above
estimates are valid for any eigenfunction but without the logarithmic factor [So88].
Moreover, these estimates are sharp without any further geometric assumptions.
Eigenfunctions Lp -estimates and their applications have been extensively studied in
the literature. For the sake of brevity we only list some relevant articles for interested
readers: [So88, So93, KoTaZw07, So11, SoZe11, SoToZe11, BlSo13, HaTa13].
Our result is in fact of particular interest in the range 2 < p ≤ 2(d+1)
, where to our
d−1
knowledge no logarithmic improvements are known. However in the range 2(d+1)
<
d−1
p < ∞, our result is weaker than the recent results of Hassell and Tacy [HaTa13]
γ(p)
where the upper bound
λj
1
(log λj ) 2
is proved for non-positively curved manifolds and
for all λj ∈ Spec∆g . We also note that, for this range of p, Sogge’s Lp bounds can be
improved by a polynomial factor in the case of the rational torus [Bo93, BoDe14].
The only former results concerning the range 2 < p ≤ 2(d+1)
which we are aware
d−1
of, are the results of Zygmund [Zy74], Bourgain [Bo93, Bo13], Sogge [So11], Sogge
and Zelditch [SoZe12b, SoZe11] and Blair and Sogge [BlSo13]. Let us recall briefly
Lp NORMS, NODAL SETS, AND QUANTUM ERGODICITY
3
the results of Blair, Sogge and Zelditch which work for geometric frameworks close
γ(p)
to ours. In [So11, BlSo13], the upper bound o(λj ) is proved for the Lp norm of
a full density subsequence of eigenfunctions when 2 < p < 2(d+1)
and the geodesic
d−1
flow is ergodic for the Liouville measure. In [SoZe12b], it is shown that this result
remains true for the L4 norm on compact surfaces satisfying a weaker dynamical
assumption; that the set of closed trajectories has zero Liouville measure on S ∗ M .
γ(p)
In [SoZe11, BlSo13], the upper bound o(λj ) is proved for the Lp norm of the
and (M, g) is nonpositively
whole sequence of eigenfunctions provided 2 < p < 2(d+1)
d−1
curved. Thus, the main improvements compared with these references are that our
method allows to treat the case of the critical exponent p = 2(d+1)
and that it
d−1
gives a logarithmic improvement on the size of the upper bound for the whole range
2 < p ≤ 2(d+1)
. On the other hand, compared with [SoZe11, BlSo13], our upper
d−1
bounds are only valid along density 1 subsequences of eigenfunctions.
Finally, we recall that the results of Zygmund and Bourgain concern the particular
2d
case of the rational torus Td = Rd /Zd . Their results shows that, for 2 < p ≤ (d−1)
,
�
the upper bound is in fact of order O(λ ) for every fixed positive �. In the case d = 2,
one can even take � = 0 [Zy74] and the upper bound remains true up to p = +∞
provided we keep � > 0 [Bo93]. In the case d = 3, the upper bound remains true up
to p = 2(d+1)
[Bo13].
d−1
Remark 1.2. Our method also gives a logarithmic improvement on L∞ bounds (See
(d−1)/4
paragraph 3.4.5). Yet, our estimates are weaker than B´erard’s upper bound
∞
λj
1
(log λj ) 2
,
[B´e77], but nevertheless it is interesting that L bounds can be improved only using
quantum ergodicity on small balls. We also note that we can obtain better Lp
estimates in the range p > 2(d+1)
, by interpolating B´erards L∞ estimate and our Lp
d−1
estimate for p = 2(d+1)
, but we omit this because it will not give us a better estimate
d−1
than those of [HaTa13].
Comments on other geometric settings. Recall that quantum ergodicity was
first proved for ergodic Hamiltonian systems in [Sh74, Ze87, CdV85, HeMaRo87].
As was already mentionned, our proof relies on a quantum ergodicity property that
holds for symbols depending only on the x variable and carried in balls of shrinking
radius. To our knowledge, this particular form of quantum ergodicity has not been
studied before except in the recent preprint by Han [Han14] (which was proved
independently) – see also [Yo13] for related results in the arithmetic setting. To
be more precise, our quantum ergodicity theorem holds for symbols supported in
balls of shrinking radius � ∼ | log λ|−K , with 0 < K < 1/(2d). Here, our logarithmic improvement in the size of the support of the symbols is the result of using
the semiclassical approximation up to the Ehrenfest time as in [Ze94, Sch06] – see
also [AnRi12] for related results.
`
HAMID HEZARI AND GABRIEL RIVIERE
4
It is natural to ask if this approach can be used in other situations where one has
equidistribution for observables depending only on the x variable. For instance,
it is known that such a property holds in the case of the torus [MaRu12, Ri13].
In [HeRi15], we show how to obtain a small scale equidistribution property in the
case of the rational torus. However, the Lp bounds we would get from this method
would not be better than the ones of [Bo13, BoDe14, Zy74]. Another interesting
case is the one of Hecke eigenfunctions on the modular surface M = P SL2 (Z)\H2
which is non-compact. On one hand, the best local Lp bounds for finite p can be
2(d+1)
obtained (as far as we know) by interpolating Sogge’s L d−1 -upper bounds with
the L∞ -upper bound from [IwSa95]. On the other hand, Luo and Sarnak also
proved a very precise rate for the quantum variance of Hecke eigenfunctions in their
quantum ergodicity theorem [LuSa95]. In particular, a direct corollary of their
result is that quantum ergodicity still holds for observables carried in shrinking
balls of radius λ−ν with ν > 0 small enough – see also [Yo13] for related results
under the Lindel¨of hypothesis . If one implements this observation in the argument
2(d+1)
of section 3 below, one would obtain a L d−1 -upper bound1 improved by a small
polynomial factor compared with Sogge’s result (for a full density subsequence of
eigenfunctions). Then, interpolating this result with the L2 norm and the L∞ upper
bounds from [IwSa95] or [Ju13], one would get upper bounds for every p ≥ 2 which
would be valid for a full density subsequence of eigenfunctions and which would
be slightly better than the ones mentionned above. However, it is plausible that
a direct application of arithmetic tools would provide better Lp upper bounds for
small range of p but we are not aware of such results.
1.2. Nodal sets. As another application of our small scale quantum ergodicity
properties, we will obtain lower bounds on the (d−1)-dimensional Hausdorﬀ measure
Hd−1 (Zψλ ) of the nodal sets.
Theorem 1.3. Let (M, g) be a smooth boundaryless compact connected Riemannian
negatively curved manifold of dimension d, and {ψj } an ONB of eigenfunctions.
Then for every δ > 0 there exists a full density subset S ⊂ N such that for every
open set U ⊂ M we have
∀j ∈ S :
Hd−1 (Zψj ∩ U ) ≥ c (log λj )
d−1
−δ
4d
3−d
λj 4 .
Here c is positive and only depends on δ and the inner radius of U . In particular,
in the 3-dimensional case we get
∀j ∈ S :
1The
1
H2 (Zψj ∩ U ) ≥ c (log λj ) 6 −δ .
Lp upper bounds are valid for compact subsets of P SL2 (Z)\H2 .
Lp NORMS, NODAL SETS, AND QUANTUM ERGODICITY
5
Background on nodal sets. For any smooth boundaryless compact connected
Riemannian manifold of dimension d, Yau’s conjecture states that there exist constants c > 0 and C > 0 independent of λ such that
√
√
c λ ≤ Hd−1 (Zψλ ) ≤ C λ.
The conjecture was proved by Donnelly and Feﬀerman [DoFe88] in the real analytic
case. In dimension 2 and in the C ∞ case, the best bounds are
√
c λ ≤ H1 (Zψλ ) ≤ Cλ3/4 .
The lower bound was proved by Br¨
uning [Br78] and Yau. The upper bound for
d = 2 was proved by Donnelly-Feﬀerman [DoFe90] and Dong [D92]. For d ≥ 3 the
existing estimates are very far from the conjecture. The best lower bound is
cλ(3−d)/4 ≤ Hd−1 (Zψλ ),
which was first proved by [CoMi11], and later by [HeSo12] and [SoZe12]. In [HeSo12]
the following slightly better estimate was proved:
√
(2)
c λ||ψλ ||2L1 ≤ Hd−1 (Zψλ ),
for any L2 -normalized eigenfunction ψλ . As was already mentionned, Blair and
Sogge obtained improvements on the Lp norms of eigenfunctions for nonpositively
curved manifolds and quantum ergodic sequences of eigenfunctions. In [BlSo13],
they recall how one can deduce from these results improved lower bounds for the L1
norm of eigenfunctions and (thanks to (2)) improvements on the results of [CoMi11,
HeSo12, SoZe12]. It is conjectured in [SoZe11] that for negatively curved manifolds
the L1 norm is bounded below by a uniform constant. In [Ze13], this conjecture
is proved under the assumption of the so called “L∞ quantum ergodicity”, which
is stronger than the standard “C 0 quantum ergodicity” and, to our knowledge, it
has not been proved or disproved in any ergodic cases. In the same article it is in
fact conjectured that L∞ quantum ergodicity holds on negatively curved manifolds.
Our logarithmic improvement on the lower bounds of the nodal sets in the case
of negatively curved manifolds is far from the above conjecture but it uses again
quantum ergodicity in a stronger sense than usual. Precisely, it uses the fact that
the eigenfunctions are still equidistributed on balls of shrinking radius � ∼ | log λ|−K ,
with 0 < K < 1/(2d). Finally, we emphasize that, compared with the above results,
our result is valid for the size of the nodal sets on any open set U inside M .
1.3. Outline of the proof. The idea of the proof of Theorem 1.3 is as follows. We
cover M with geodesic balls B of radius � = (log λ)−K . We then fix such a ball B and
use our refined version of quantum ergodicity to obtain a full density subsequence
of eigenfunctions which take L2 mass on B comparable to vol(B). Then, we apply
the method of [CoMi11] and obtain a lower bound for the size of the nodal set
in this shrinking ball. This latter property requires to control the Lp norm of the
eigenfunctions inside these shrinking ballls. For that purpose, we rescale our problem
and make use of semiclassical Lp estimates for quasimodes [So88, KoTaZw07, Zw12].
6
`
HAMID HEZARI AND GABRIEL RIVIERE
As a corollary of this intermediary step, we also get the proof of Theorem 1.1.
Finally, we conclude the proof of Theorem 1.3 by adding up over all balls in the
covering. We observe that, since the number of balls is �−d one has to be careful in
choosing a full density subsequence that works uniformly for all balls in the covering.
Remark 1.4. After the initial posting of this article, Sogge in [So15] was able to
improve our arguments in paragraph 3.3 in order to obtain localized Lp estimates.
These estimates establish a clear relation between Lp bounds of eigenfunctions and
estimating L2 mass of eigenfunctions in balls with shrinking radius. We refer to his
note for a precise statement.
The article is organized as follows. In Section 2, we prove our refined version of
quantum ergodicity on negatively curved manifolds, and then, in Section 3, we
apply these results to the study of Lp norms and nodal sets. In Appendix A, we
give the proof of a dynamical result on the rate of convergence of Birkhoﬀ averages
with respect to the Lp norm.
Remark 1.5. Since we are going to use semiclassical techniques we prefer to use the
semiclassical parameter � and we will use this for the rest of this paper. For readers
√
who are not familiar with the semiclassical parameter �, one can think of � as 1/ λ.
Hence the high energy regime λ → ∞ is equivalent to the semiclassical limit � → 0.
2. Quantum ergodicity on small scale balls
In all of this section, M is a smooth, connected, compact Riemannian manifold
without boundary and with negative sectional curvature.
Let α > 0 be some fixed positive number. For every 0 < � ≤ 1, we consider an
orthonormal basis (made of eigenfunctions of −�2 ∆g ) (ψ�j )j=1,...,N (�) of the subspace
H� := 1[1−α�,1+α�] (−�2 ∆g )L2 (M ).
According to the Weyl’s law for negatively curved manifolds (see [DuGu75, B´e77]),
one has N (�) ∼ αAM �1−d for some constant AM depending only on (M, g). For each
1 ≤ j ≤ N (�), we denote Ej (�) ∈ [1−α�, 1+α�] to be the eigenvalue corresponding
to ψj� :
−�2 ∆ψ�j = Ej (�)ψ�j .
Let 0 ≤ χ ≤ be a smooth cutoﬀ function which is equal to 1 on [−1, 1] and to 0
outside [−2, 2]. For a given x0 in M and a 0 < � < inj(M,g)
, we define
10
�
�
� exp−1
x0 (x)�x0
χx0 ,� (x) = χ
,
�
Lp NORMS, NODAL SETS, AND QUANTUM ERGODICITY
7
where expx0 is the exponential map associated to the metric g, and �v�2x0 = gx0 (v, v).
By construction, this function is compactly supported in B(x0 , 2�). The purpose of
this section is to describe the rate of convergence of the following quantity:
�2p
�
N (�) ��
�
1 � ��
j 2
V�,p (x0 , �) :=
χx0 ,� |ψ� | dvg −
χx0 ,� dvg �� ,
�
N (�) j=1 M
M
where p is a fixed positive integer and vg is the normalized volume measure on M .
Observe that there exists uniform (in x0 and in �) constants 0 < c˜1 < c˜2 such that
�
d
(3)
c˜1 � ≤
χx0 ,� dvg ≤ c˜2 �d .
M
It follows from (3) that to obtain some nontrivial rate we need to show that V�,p (x0 , �)
is at least o(�2dp ). We will show that this is in fact possible for a particular choice
of � as a function of �.
Our main result on the rate of convergence of V�,p (x, �) to zero is the following:
Proposition 2.1. Let M be a smooth, connected, compact Riemannian manifold
1
without boundary and with negative sectional curvature. Let 0 < K < 2d
, let m > 0
and β > 0.
Then, there exists C > 0 and �0 > 0 such that the following holds, for every 0 <
� ≤ �0 :
Given an orthonormal basis (ψ�j )j=1,...,N (�) of the subspace H� made of eigenfunctions
of −�2 ∆g , one has, for every x0 ∈ M ,
C
V�,p (x0 , m| log �|−K ) ≤
.
| log �|p(1−2Kβ)
The assumption 0 < K < 1/(2d) implies that the upper bound is (at least for β > 0
small enough) o(�2dp ) as expected. We point out that the main results of the recent
preprint [Han14] give a small scale quantum ergodicity result (with p = 1) that
holds for more general classes of symbols.
2.1. Classical ergodicity. Before giving the proof of Proposition 2.1, we need some
precise information on dynamics of the geodesic flow of negatively curved manifolds.
Let 0 < β < 1, and denote by C β (S ∗ M ) the space of β-H¨older functions on S ∗ M .
Then, one has the following property of exponential decay of correlations [Li04]:
Theorem 2.2. Let M be a smooth, connected, compact Riemannian manifold without boundary and with negative sectional curvature. Let 0 < β < 1.
Then, there exists some constants Cβ > 0 and σβ > 0 such that, for every a and b
in C β (S ∗ M ), and for every t in R,
��
�
�
�
�
�
t
−σβ |t|
�
�
�a�C β �b�C β ,
� ∗ (a ◦ G )bdL − ∗ adL ∗ bdL� ≤ Cβ e
S M
S M
S M
8
`
HAMID HEZARI AND GABRIEL RIVIERE
where L is the normalized Liouville measure on S ∗ M , Gt is the geodesic flow, and
�.�C β is the H¨older norm.
This result on the rate of mixing has implications on the rate of classical ergodicity.
More precisely, given a in C β (S ∗ M, R) and T > 0, one can define, for every positive
integer p,
� � T
�2p
�
�
�1
�
t
�
� dL.
Vp (a, T ) :=
a
◦
G
dt
−
adL
�
�
S∗M T 0
S∗M
Cβ,1 �a�2
Cβ
Using Theorem 2.2, one can easily show that V1 (a, T ) ≤
. Moreover, we can
T
also control the moments of order 2p but obtaining such estimates requires more
work:
Proposition 2.3. Let M be a smooth, connected, compact Riemannian manifold
without boundary and with negative sectional curvature. Let 0 < β < 1, and let p
be a positive integer. Then, there exists some constant Cβ,p > 0 such that, for all
symbols a in C β (S ∗ M ), and for every T > 0,
� � T
�2p
�
�
2p
�1
�
t
�
� dL ≤ Cβ,p �a�C β .
Vp (a, T ) :=
a
◦
G
dt
−
adL
�
�
Tp
S∗M T 0
S∗M
A slightly weaker version of this inequality can be obtained as a corollary of the
central limit theorem proved in [Ze94, MeTo12]. In both references, the proof was
based on results due to Ratner [Ra73], and the dependence of the constant in terms
of H¨older norms was not very explicit. In Appendix A, we will use a slightly different approach which does not require the use of Markov partitions and symbolic
dynamics, and which gives the above explicit constants in terms of the H¨orlder norm
of the symbol. More precisely, we will use the machinery developed (for instance) by
Liverani in [Li04] in order to study the spectral properties of the transfer operator.
We emphasize that, unlike the case p = 1, the rate of convergence of higher moments
cannot be directly deduced from Theorem 2.2, and that we really need to use more
precise results from [Li04] in order to get Proposition 2.3.
In the case where we pick the function χx,� defined above, we get
� � T
�2p
�
�
�
�
�1
�
1
t
�
(4)
Vp (χx,� , T ) :=
χx,� ◦ G dt −
χx,� dvg �� dL = O 2pβ p ,
�
� T
S∗M T 0
M
where the constant in the remainder can be chosen uniformly in terms of x, � and
T . We observe that this is valid for any β > 0 (of course the constant depends on
β).
2.2. Proof of Proposition 2.1. We will now implement these dynamical properties in the “standard” proof of quantum ergodicity [Sh74, Ze87, CdV85, HeMaRo87,
Lp NORMS, NODAL SETS, AND QUANTUM ERGODICITY
9
Ze96, Zw12]. As in [Ze94, Sch06], we will make use of the semiclassical approximation up to times of order | log �|.
Let x be a point in M , let 0 < K <
1
2d
and let m > 0. We define
� = �� := m| log �|−K .
�
We set χx,� := χx,� − M χx,� dvg . In particular, for every 0 < ν < 12 , χx,� belongs to
the “nice” class of symbols Sν0,0 (T ∗ M ) as defined in [Zw12] for instance. This class
of symbols is suitable to pseudodiﬀerential calculus (composition laws, Weyl’s law,
Egorov’s theorem, etc.).
Remark 2.4. We introduce 0 ≤ χ1 ≤ 1 a smooth cutoﬀ function which is identically
1 in a small neighborhood of 1, and which vanishes outside a slightly bigger neighborhood. We observe that, for � > 0 small enough, one has χ1 (−�2 ∆g )ψ�j = ψ�j
for every 1 ≤ j ≤ N (�). We also recall that χ1 (−�2 ∆g ) is a �-pseudodiﬀerential
operator in Ψ−∞,0 (M ) with principal symbol θ(x, ξ) := χ1 ((�ξ�∗x )2 ) [Zw12] (chapter
14). We have denoted by �.�∗x the induced metric on Tx∗ M .
We then get
N (�)
��2p
1 � �� j
V�,p (x, �) =
ψ� , Op� (θχx,� )ψ�j � + O(�),
N (�) j=1
where Op� is a fixed quantization procedure [Zw12]. Without loss of generality, we
can suppose that Op� (1) = IdL2 (M ) and that it is a positive quantization procedure [Ze87, CdV85, HeMaRo87, Ze94], i.e. Op� (a) ≥ 0 if a ≥ 0. We now let
T (�) = κ0 | log �|,
�
�
with κ0 > 0. Since ψ�j are ∆g -eigenfunctions, we can replace ψ�j , Op� (θχx,� )ψ�j by
� T (�)
� j −it�∆g /2
�
1
ψ� , e
Op� (θχx,� )eit�∆g /2 ψ�j dt.
T (�) 0
We now choose and fix κ0 small enough (only depending on (M, g)) and apply the
Egorov theorem up to T (�)(see [Zw12], section 11.4) to find that
�
� ��2p
�
N (�) ��
�
1 � �� j
1 T
t
V�,p (x, �) =
ψ� , Op� θ
χx,� ◦ G dt ψ�j �� + O(�ν0 ),
�
N (�) j=1
T 0
for some fixed ν0 > 0 (depending on κ0 > 0). Moreover, still thanks to the Egorov
�T
theorem, the principal symbol T1 0 χx,� ◦ Gt dt belongs to Sν0,0 (T ∗ M ) for some 0 <
ν < 1/2. Because Op� is a positive quantization procedure, the distribution
�
�
µj� : b ∈ Cc∞ (T ∗ M ) �→ ψ�j , Op� (b) ψ�j
`
HAMID HEZARI AND GABRIEL RIVIERE
10
is positive on T ∗ M , and in fact it is a positive measure on T ∗ M . Then by applying
the Jensen’s inequality (recall that p ≥ 1) we get
�
��
�2p � �
� T
N (�)
�
�
1 �
1
V�,p (x, �) =
ψ�j , Op� ��θ
χx,� ◦ Gt dt��
ψ�j + O(�ν0 ).
N (�) j=1
T 0
We now apply the local Weyl law (see for instance Proposition 1 in [Sch06] for a
proof of this fact in our context). We get
�2p
�
�
�
1 T
t
V�,p (x, �) ≤ C0
θ
χx,� ◦ G dt
dL + O(�ν0 ),
T
∗
T M
0
where C0 is independent of � and x in M . The parameter ν0 > 0 can become smaller
from line to line but it remains positive. We underline that the diﬀerent constants
do not depend on the choice of the ONB of H� .
Remark 2.5. We also emphasize that, up to this point, all the constants in the
remainder are uniform for x in M (by construction of the function χx,� ). Precisely,
they only depend on a finite number derivatives of χ, the manifold (M, g), and the
choice of the quantization procedure.
We now apply property (4), and we get that2
�
�
1
V�,p (x, �) ≤ O 2βp
+ O(�ν0 ),
� | log �|p
which implies Proposition 2.1.
3. Proof of the main results
Again, even if we do not mention it at every step, in this section M will be a smooth,
connected, compact Riemannian manifold without boundary and with negative
sectional curvature.
In this section we give the proofs of our main results. The strategy is as follows.
We start by covering the manifold with balls of radius � = | log �|−K , and we apply
our quantum ergodicity result from the previous section to extract a density 1 subsequence “adapted” to this cover of M . We then improve the Lp bounds along this
subsequence of eigenfunctions by rescaling our problem and using the semiclassical
Lp estimates for quasimodes [So88, KoTaZw07, Zw12]. After that, we use the approach of Colding-Minicozzi [CoMi11] to prove our lower bound for the size of the
nodal set on any open subset U .
2Recall
that T = T (�) = κ0 | log �|.
Lp NORMS, NODAL SETS, AND QUANTUM ERGODICITY
11
3.1. Covering M with small balls. First, we cover M with balls (B(xk , �))k=1,...,R(�)
1
of radius � = | log �|−K , for a fixed 0 < K < 2d
. In addition, we require that the
covering is chosen in such a way that each point in M is contained in Cg -many of
the double balls B(xk , 2�). The number Cg can be chosen to be only dependent on
(M, g), and hence independent of 0 < � ≤ 1. See for example Lemma 2 in [CoMi11]
for a proof of this fact. We have
R(�)
M⊂
�
B(xk , �).
k=1
The doubling property shows that there exists constants c1 > 0 and c2 > 0 , only
dependent on (M, g), such that
(5)
c1 �−d ≤ R(�) ≤ c2 �−d .
We note that in fact we have a family of coverings parametrized by �, and hence the
centers {xk } also depend on �. Now let 1 ≤ k ≤ R(�). As in section 2, we define
�
�
� exp−1
xk (x)�xk
χxk ,� (x) = χ
,
�
where expxk is the exponential map associated to the metric g. By construction,
this function is compactly supported in B(xk , 2�).
3.2. Extracting a full density subsequence. Following the proof of [Zw12], we
want to extract a subsequence of density 1 of eigenfunctions for which quantum
ergodicity holds for all symbols χxk ,� uniformly in k. This is a standard procedure,
and we just have to pay attention to the fact that the number of balls in the cover
is large.
1
Let m = 12 or m = 50. Let 1 ≤ k ≤ R(�), 0 < K < 2d
, and β > 0 such that
1
K < 2d+4β . We also fix a positive integer p such that dK + p(K(4β + 2d) − 1) < 0.
As before, we fix
� = �� := | log �|−K .
Combining the Bienaym´e-Tchebychev inequality to Proposition 2.1 and to inequality (3), the subset Jk,K,m (�) ⊂ {1, . . . , N (�)} defined as
��
�
�
�
j 2
�
�
� M�χxk ,m� |ψ� | dvg
�
(6)
Jk,K,m (�) := j ∈ {1, . . . , N (�)} : �
− 1 � ≥ �β ,
�
�
χ
dv
g
M xk ,m�
satisfies
(7)
�Jk,K,m (�)
C0
≤ (4β+2d)p
,
N (�)
�
| log �|p
12
`
HAMID HEZARI AND GABRIEL RIVIERE
where the positive constant C0 depends only on p, on χ, and on the manifold (M, g).
In particular, C0 is uniform for 1 ≤ k ≤ R(�). We then define
and
Λk,K,m (�) := {1 ≤ j ≤ N (�) : j ∈
/ Jk,K,m (�)},
R(�)
ΛK,m (�) :=
�
Λk,K,m (�).
k=1
Thanks to (5) and to (7), we get
R(�)
�ΛK,m (�)
1 �
C 0 c2
≥1−
�Jk,K,m (�) ≥ 1 − (4β+2d)p+d
.
N (�)
N (�) k=1
�
| log �|p
�Λ
(�)
K,m
Hence using the definition of � = �� , we find that N
tends to 1 as � goes to
(�)
0. This in particular, using (6) and (3), shows that � > 0 small enough, for every
1 ≤ k ≤ R(�), and for every j ∈ ΛK,m (�),
�
d
a1 � ≤
χxk ,m� |ψ�j |2 dvg ≤ a2 �d ,
M
where the constants a1 , a2 > 0 depend only on χ and on the manifold (M, g).
By taking intersection ΛK (�) := ΛK,1/2 (�) ∩ ΛK,50 (�), we still have a full density
subsequence. Hence we have proved the following crucial lemma
1
Lemma 3.1. Suppose that M is negatively curved. Let 0 < K < 2d
and � =
−K
| log �| . There exists 0 < �0 ≤ 1/2 such that, for every 0 < � ≤ �0 , the following
holds:
Given an orthonormal basis (ψ�j )j=1,...N (�) of 1[1−α�,1+α�] (−�2 ∆g )L2 (M ) made of
eigenfunctions of −�2 ∆g , one can find a full density subset ΛK (�) of {1, . . . , N (�)}
such that, for every 1 ≤ k ≤ R(�), and for every j ∈ ΛK (�), one has
�
�
j 2
d
(8)
a1 � ≤
|ψ� | dvg ≤
|ψ�j |2 dvg ≤ a2 �d ,
B(xk ,�)
B(xk ,50�)
where the constants a1 , a2 > 0 depend only on χ and on the manifold (M, g).
This lemma is the key ingredient that we will use to improve the usual Sogge Lp
upper bounds on eigenfunctions and the lower bounds on the size of nodal sets along
the full density subset.
3.3. Lp -estimates. In this part we make use of semiclassical Lp estimates for quasimodes as in [Zw12] (Chapter 10) to obtain our new Lp estimates. In fact, we will
prove something slightly stronger than what we stated in the introduction and show
that the Lp norm of the eigenfunctions in the balls B(xk , �) can be controlled. In
the proof we will make use of our quantum ergodicity property (8). The key result
of this section is the following:
Lp NORMS, NODAL SETS, AND QUANTUM ERGODICITY
Proposition 3.2. Let p =
2(d+1)
d−1
13
and 0 < K < 1/(2d). Fix
� = | log �|−K .
Then, there exists C0 > 0 (depending only on (M, g), K and χ) such that, for every
1 ≤ k ≤ R(�) and for every j ∈ ΛK (�), one has
�
1
�
|ψ�j |p dvg ≤ C0 .
d
� B(xk ,2�)
�
Using (5), we easily deduce our main result on the Lp norm of eigenfunctions of the
Laplacian. Namely, there exists CK such that for all j ∈ ΛK (�)
� 1
�
K
2(d + 1)
�ψ�j �Lp (M ) ≤ CK �− p | log �|− p , p =
.
d−1
We can then use an interpolation with L2 norm to get our result for all 2 < p <
2(d+1)
.
d−1
3.4. Proof of Proposition 3.2. In order to prove this Proposition, we will first
draw a few consequences of our quantum ergodicity result. We will then show that,
along ΛK (�), and after rescaling using the change of coordinates x = expxk (�y), the
restriction of the eigenfunctions to the rescaled balls are quasimodes of order
h = �/�,
(9)
of a certain h-pseudodiﬀerential operator (the rescale of ∆g ). Finally we will use
the semiclassical Lp estimates from [Zw12] to establish an upper bound for the Lp
norm of these rescaled quasimodes.
3.4.1. Preliminary observations. Let 1 ≤ k ≤ R(�). Clearly
�
�
j p
|ψ� | dvg ≤
|χxk ,2� ψ�j |p dvg .
B(xk ,2�)
B(xk ,4�)
Using the change of coordinates x = expxk (�y), we get
�
�
j p
d
(10)
|ψ� | dvg ≤ CM �
|χ(�y�xk /2)ψ�j ◦ expxk (�y)|p dy,
B(xk ,2�)
Bk (0,4)
where Bk (0, 4) is the ball centered at 0 in Txk M and CM depends only on the
manifold (M, g). We will now consider the following element of L2 (Bk (0, 4)):
u˜j�,� (y) := χ(�y�xk /2)ψ�j ◦ expxk (�y),
and show that it is a quasimode of a certain order for a “rescaling” of the operator
−�2 ∆.
14
`
HAMID HEZARI AND GABRIEL RIVIERE
Before that, we draw two simple consequences of the quantum ergodicity property.
The first observation is that by (8), we get
(11)
� j
�
�ψ ◦ expx (�y)� 2
�
k
L (B
k (0,50))
≤ C M a2 ≤ c M
�
a2 �
�u˜j �
,
a1 �,� L2 (Rd )
where CM and cM depends only on (M, g). The second observation is that clearly
the first inequality also implies that
� j �
�u˜ �
(12)
�,� L2 (Rd ) = O(1).
3.4.2. Rescaled Laplacian. We now introduce the following diﬀerential operator on
T xk M � R d :
�
�
2
�
�
�
∂
1
∂
∂
Qh := −h2 χ(�y�xk /20)
g i,j (�y)
+
(Dg g i,j )(�y)
,
∂y
∂y
D
(�y)
∂y
∂yj
i
j
g
i
i,j
where (g i,j ) are the matrix
� coeﬃcients of the metric g �in the coordinates y = κk (x) :=
exp−1
(x),
and
D
=
det(gi,j ). We recall that h = � .
g
xk
Since ψ�j is an eigenmode of −�2 ∆g with eigenvalue E in [1 − α�, 1 + α�], clearly
one has
�
�
j
−1
∀y ∈ Bk (0, 4), Qh ψ�j ◦ κ−1
k (�y) = Eψ� ◦ κk (�y).
This implies that
�
�
�
�
j �
�(Qh − E)˜
uj�,� �L2 (Rd ) = �[Qh , χ(�.�xk /2)]ψ�,�
,
L2 (Rd )
j
where ψ�,�
(y) := χ(�y�xk /4)ψ�j ◦ κ−1
k (�y). Using the composition laws for pseudodifd
ferential operators in R , we get
w
2
[Qh , χ(�.�xk /2)] = h Opw
h (r0 ) + h Oph (r1 ),
d
Opw
r1 ∈ S 0,0 (R2d ) and r0 := 1i {q0 , χ(�.�xk /2)}
h denotes the Weyl quantization on R ,�
in S 1,0 (R2d ) with q0 (y, η) = χ(�y�xk /20) i,j g i,j (�y)ηi ηj . Thanks to the Calder´onVaillancourt theorem, we get
�
�
�
�
�
�
j �
2� j �
�(Qh − E)˜
uj�,� �L2 (Rd ) ≤ h �Opw
ψ�,� L2 (B (0,50)) ,
h (r0 )ψ�,� L2 (Rd ) + Ch
k
for some uniform constant C > 0. Therefore by (11), we get
�
�
�
�
� �
j �
2 � j �
�(Qh − E)˜
(13)
uj�,� �L2 (Rd ) ≤ h �Opw
˜�,� L2 (Rd ) .
h (r0 )ψ�,� L2 (Rd ) + O(h ) u
Lp NORMS, NODAL SETS, AND QUANTUM ERGODICITY
15
3.4.3. Order of the quasimode. In order to show that u˜j�,� is in fact a O(h) quasimode
�
�
j �
for the operator Qh , by (13) it just remains to show that �Opw
(r
)ψ
0
h
�,� L2 (Rd ) is
w
uniformly bounded. The only diﬃculty is that Oph (r0 ) is not a priori bounded on
L2 as r0 belongs to S 1,0 (R2d ). However we can easily overcome this issue by inserting
an appropriate smooth cutoﬀ function in ξ variable as follows.
First, we write
(14)
� −1 ∗
�
j
j
w ˜
Opw
h (r0 )ψ�,� (y) = T� Op� (T� (r0 )) (κk ) (χxk ,4� ψ� ) ,
where T˜� (r0 )(y, η) := r0 (y/�, η), and T� (f )(y) := f (�y). Hence
�
� w
�
� −1 ∗
�
˜
�Oph (r0 )ψ j �
(15)
= �−d/2 �Opw
� (T� (r0 )) (κ ) (χx
�,� L2 (Rd )
k
��
j �
ψ
)
�
k ,4� �
L2 (Rd )
.
Using Remark 2.4, we observe that χ1 (−�2 ∆g )ψ�j = ψ�j , and that χ1 (−�2 ∆g ) is an element of Ψ−∞ (M ) with principal symbol θ(x, ξ) = χ1 ((�ξ�∗x )2 ) ([Zw12], Chapter 14).
∗
Therefore, by replacing ψ�j with χ1 (−�2 ∆g )ψ�j and noting that T˜� (r0 )(κ−1
k ) (χxk ,� θ)
2d
belongs to S(R ), we obtain
�
�
�
� −1 ∗
��
� w ˜
j �
Op
(
T
(r
))
(κ
)
(χ
ψ
)
≤ O(1) �χxk ,10� ψ�j �L2 (M ) + O(�1−2ν ).
� � � 0
xk ,4� � �
k
2
d
L (R )
where 0 < ν < 1/2 is such that �ν ≤ � = �� for � small enough and where the
constants are uniform for 1 ≤ k ≤ R(�). Applying this upper bound to (15), and
using Lemma 3.1, and the fact that � = | log �|−K , we get
� w
�
�Oph (r0 )ψ j �
(16)
�,� L2 (Rd ) ≤ O(1).
d
Remark 3.3. We note that, in order to get a O(1), we need that �1−2ν �− 2 remains
bounded. As we will take � = | log �|−K , this restriction does not matter.
Finally, combining this with (13) and (11), we arrive at
�
�
� �
�(Qh − E)˜
uj�,� �L2 (Rd ) = O(h) �u˜j�,� �L2 (Rd ) ,
where the constant in the remainder is uniform for j in ΛK (�) and for 1 ≤ k ≤ R(�)
(note that the rescaled quasimode u
˜j�,� depends on k). This is precisely the definition
of a quasimode of order h.
3.4.4. Semiclassical Lp -estimates. We are now in position to apply the semiclassical Lp estimates from [Zw12] (precisely Theorem 10.10), which are valid for O(h)quasimodes.
Using Remark 2.4 and similar arguments as we delivered in paragraph 3.4.3, we can
see that the sequence (˜
uj�,� )0<h≤h0 is localized in a compact subset K of phase space
`
HAMID HEZARI AND GABRIEL RIVIERE
16
in the sense of section 8.4 of [Zw12]. For instance, we can take K to be
1
Bk (0, 5) × { ≤ ξ 2 ≤ 3/2}.
2
We then note that the principal symbol
q0 − E = χ(�y�2xk /20)
of Qh − E satisfies:
�
g jk (�y)ηj ηk − E,
• ∂η (q0 − E) �= 0 on {q0 = E} ∩ K,
• for all x0 ∈ Bk (0, 5), the hypersurface {q0 (x0 , ξ) = E} has a nonzero second
fundamental form (nonzero curvature).
Hence all the required conditions of Theorem 10.10 in [Zw12], for the semiclassical
Lp estimates to hold, are satisfied and we get
�˜
uj�,� �Lp (Rd ) ≤ Ch−1/p �˜
uj�,� �L2 (Rd ) ,
p=
2(d + 1)
,
d−1
where C only depends on (M, g), and on a finite number of derivatives of χ, χ1 , and
the exponential map. Therefore, by (10) and (12), we get
� �1
j
− dp
� � p
� �ψ� �Lp (B(xk ,2�)) ≤ C
,
�
where the constant is uniform for j ∈ ΛK (�) and 1 ≤ k ≤ R(�). We recall again
that h = ��
3.4.5. L∞ bounds. Using the same rescaling argument as in the last section and also
the semiclassical L∞ estimates of [KoTaZw07] and [SmZw13], we get
�ψ�j �L∞ (B(xk ,2�)) ≤ C�
d−1
2
�−
d−1
2
,
where C is again independent of k and j ∈ ΛK (�). We then have
�ψ�j �L∞ (M ) = max{�ψ�j �L∞ (B(xk ,2�)) } ≤ C�
k
d−1
2
�−
d−1
2
.
This proves our claimed L∞ bounds.
3.5. Lower bounds for nodal sets on a fixed open set U . Recall that we want
to give a lower bound on the size of nodal sets on every fixed open set U . If we were
only interested in the case U = M , then we could conclude more quickly using the
results of [SoZe11, HeSo12] (see paragraph 3.5.4).
Lp NORMS, NODAL SETS, AND QUANTUM ERGODICITY
17
3.5.1. Colding-Minicozzi approach. The main lines of our argument are similar to
the ones in [CoMi11]. The proof in this reference was based on the study of the
so-called q-good balls. Namely, given a constant q > 1 and an eigenmode ψ� of the
operator −�2 ∆g , a ball B(x, r) ⊂ M (centered at x and of radius r > 0) is said to
be q-good if
�
�
B(x,2r)
|ψ� |2 dvg ≤ 2q
B(x,r)
|ψ� |2 dvg .
A ball which is not q-good is said to be q-bad. The main advantage of q-good balls
is that you can control from below the volume of the nodal set on them. Precisely,
it was shown in [CoMi11] (Proposition 1) that
Proposition 3.4. Let q > 1 and let r0 > 1. There exist µ > 0 and �0 > 0 so that
if 0 < � ≤ �0 , E ∈ [1 − α�, 1 + α�], −�2 ∆g ψ� = Eψ� on B(x, r) ⊂ M with r ≤ r0 �,
ψ� vanishes somewhere on B(x, r/3), and
�
�
2
q
|ψ� | dvg ≤ 2
|ψ� |2 dvg ,
B(x,2r)
B(x,r)
then
Hn−1 (B(x, r) ∩ {ψ� = 0}) ≥ µrn−1 .
Remark 3.5. We stress that this proposition is valid for any smooth compact Riemannian manifold without any assumption on the curvature.
This proposition provides a local estimate on the nodal set provided that we are on a
q-good ball on which ψ� vanishes. We now state a result of Courant which guarantees
such a vanishing property (for a proof see for example Lemma 1 of [CoMi11])
Lemma 3.6. There exists r0 > 0 so that if 0 < � ≤ 1/2, E ∈ [1 − α�, 1 + α�],
−�2 ∆g ψ� = Eψ� , then ψ� has a zero in every ball of radius r03� .
By applying these two properties to a proper cover of M with balls of radius r0 �,
the proof of [CoMi11] boils down to counting the number of q-good balls in a cover
(for some large enough q > 0).
Our plan is to use a similar counting argument as [CoMi11] except that we will take
advantage of the small-scale quantum ergodic property of eigenfunctions. For such
eigenfunctions, we will be able to count the number of q-good balls in balls of radius
� of order | log �|−K (with 0 < K < 1/(2d)).
3.5.2. Lower bounds on the number of q-good balls. First, we cover M with balls
(B(˜
xk , r0 �))k=1,...,R(�)
with r0 defined in Lemma 3.6. In particular, any eigenfunc˜
2
tion ψ� of −� ∆g on the space 1[1−α�,1+α�] (−�2 ∆g )L2 (M ) has a zero in every ball
B(˜
xk , r0 �/3). In addition, we require that the covering is chosen in such a way
`
HAMID HEZARI AND GABRIEL RIVIERE
18
that each point in M is contained in Cg -many of the double balls B(˜
xk , 2r0 �). Recall that the number Cg can be chosen to be only dependent on (M, g), and hence
independent of 0 < � ≤ 1/2 [CoMi11].
We now fix 1 ≤ k ≤ R(�) and estimate the number of q-good balls 3 inside B(xk , 2�).
Let {B(˜
xl , r0 �) : l ∈ Ak } be the set of those balls in the covering which have
non-empty intersection with B(xk , �). Then, taking � > 0 small enough to ensure
4r0 � ≤ �, one has
�
�
B(xk , �) ⊂
B(˜
xl , r0 �) ⊂
B(˜
xl , 2r0 �) ⊂ B(xk , 2�).
l∈Ak
l∈Ak
We denote by Akq−good the subset�of indices such that l ∈ Ak and B(˜
xl , r0 �) is a
q-good ball. Then, we let Gk = l∈Ak
B(˜
xl , r0 �). We start our proof with the
q−good
following lemma which is an analogue of Lemma 3 of [CoMi11] for balls of radius �:
Lemma 3.7. There exists qM > 1 depending only on (M, g) such that, for every
� > 0 small enough, and, for every j ∈ ΛK (�), one has
�
1
(17)
|ψ�j |2 dvg ≥ a1 �d ,
2
Gk
where a1 > 0 is the same constant as in Lemma 3.1.
Proof. To show this we first note that, using Lemma 3.1,
�
�
�
�
j 2
j 2
j 2
d
|ψ� | dvg ≥
|ψ� | dvg −
|ψ� | dvg ≥ a1 � −
Gk
Bk
B(xk ,�)
�
Bk
|ψ�j |2 dvg ,
where Bk = l∈Ak −Ak
B(˜
xl , r0 �). On the other hand, using Lemma 3.1 one more
q−good
time, one has
�
�
�
j 2
|ψ� | dvg ≤
|ψ�j |2 dvg
Bk
l∈Ak −Akq−good
<2
−q
B(˜
xl ,r0 �)
�
l∈Ak −Akq−good
≤ Cg 2
−q
�
B(xk ,2�)
�
B(˜
xl ,2r0 �)
|ψ�j |2 dvg
|ψ�j |2 dvg
−q d
≤ C g a2 2 � .
3We
underline that we count the number of q-good balls for a given eigenfunction in the full
density subsequence obtained in Lemma 3.1. Also in [CoMi11], the authors count the number of
q-good balls over M .
Lp NORMS, NODAL SETS, AND QUANTUM ERGODICITY
19
Therefore, by choosing q large enough so that Cg a2 2−q ≤ 12 a1 , we have proved (17).
�
To estimate the number of q-good balls which intersect B(xk , �), we will estimate
vol(Gk ). Unfortunately, since Gk is a complicated set and, since it depends on � in
a way which is hardly tractable, we cannot use quantum ergodicity to get a lower
bound for vol(Gk ). Instead, we will use our Lp estimates 4 from Proposition 3.2.
First, we write H¨older’s inequality with p = 2(d+1)
to get
(d−1)
�
�
� p−2
1 d
(18)
a1 � ≤
|ψ�j |2 dvg ≤ vol(Gk ) p ||ψ�j ||2Lp (B(xk ,2�)) .
2
Gk
By applying Proposition 3.2 to (18), we find the following lower bound on the volume
of Gk :
d+1
d−1
vol(Gk ) ≥ a0 � 2 � 2 ,
for some positive constant a0 that depends only on (M, g) and χ. This implies that
�Akq−good ≥ a
˜0 �
(19)
d+1
2
�−
d+1
2
,
where a
˜0 > 0 and depends only on (M, g) and χ.
3.5.3. Conclusion. By combining Proposition 3.4 with the lower bound (19), we find
˜
that for every 0 < � ≤ �0 , for every 1 ≤ k ≤ R(�),
and for every j ∈ ΛK (�), we have
(20)
� � � d+1
�
�
�
�
�
2
d−1
d−1
d−1
c0
µ�
≤
H
Zψj ∩ B(˜
xl , r0 �) ≤ Cg H
Zψj ∩ B(xk , 2�) ,
�
�
�
k
l∈Aq−good
To finish the proof of Theorem 1, let Ur be an arbitrary open ball of radius r
(independent of �), and Ur/2 be a ball of radius r/2 concentric to Ur . Also let
{B(xk , �) : k ∈ A} be the balls with nonempty intersection with U r2 . Clearly
�
U r2 ⊂
B(xk , 2�) ⊂ Ur ,
k∈A
and A must have at least c�−d elements (for some uniform constant c > 0 depending
on r). Hence using (20),
Hd−1 (ZψEj (�) ∩ Ur ) ≥ C�
Since 0 < K <
4Sogge’s
1
,
2d
d−3
2
| log �|
K(d−1)
2
.
we get the claimed range of exponents for our logarithmic factor.
Lp estimates were already used in the proof of [CoMi11], but here we use our slightly
stronger version.
`
HAMID HEZARI AND GABRIEL RIVIERE
20
3.5.4. Alternative proof for U = M . When U = M , we can conclude more quickly
using inequality (2) from the introduction. In fact, thanks to the H¨older’s inequality,
one has
1
� �1
−1
1 = �ψ�j �Lθ 2 ≤ �ψ�j �L1 (M ) �ψ�j �Lθ p (M ) ,
p−2
where p = 2(d+1)
and θ = 2(p−1)
. This inequality was already used in [SoZe11] (end
d−1
of paragraph 1.1). Hence by applying our improved Lp estimates, we find that
C1 �
d−1
4
| log �|
K(d−1)
4
≤ �ψ�j �L1 (M ) ,
for some uniform constant C1 > 0. The conclusion now follows from inequality (2)
1
and the fact that we can allow any exponent 0 < K < 2d
.
Appendix A. Moments of order 2p
In this appendix, we give the proof of Proposition 2.3. Before that, we need to recall
a few results and notations from [Li04] – see paragraphs A.1 and A.2. We deduce
from these properties a multicorrelation result in paragraph A.3 which generalizes
Theorem 2.2. Finally, using an argument similar to Lemma 3.2 of [Ra73], we deduce
Proposition 2.3.
A.1. Anosov property. When M is a negatively curved manifold, the geodesic
flow5 (Gt )t∈R satisfies the Anosov property on S ∗ M . Precisely, it means that, for
every ρ = (x, ξ) in S ∗ M , there exists a Gt -invariant splitting
(21)
∗
Tρ T1/2
M = RX0 (ρ) ⊕ E s (ρ) ⊕ E u (ρ),
∗ 2
where X0 (ρ) is the Hamiltonian vector field associated to p0 (x, ξ) = (�ξ�2x ) , E u (ρ) is
the unstable direction and E s (ρ) is the stable direction. These three subspaces are
preserved under the geodesic flow and there exist constants C0 > 0 and γ0 > 0 such
that, for any t ≥ 0, for any v s ∈ E s (ρ) and any v u ∈ E u (ρ),
u
�dρ Gt0 v s �Gt0 (ρ) ≤ C0 e−γ0 t �v s �ρ and �dρ G−t
≤ C0 e−γ0 t �v u �ρ ,
0 v �G−t
0 (ρ)
where �.�w is the norm associated to the Sasaki metric on S ∗ M .
A.2. Banach spaces adapted to the transfer operator. In this setting, one can
study the spectral properties of the transfer operator which is defined as follows:
∀f ∈ C ∞ (S ∗ M ), Lt f = f ◦ G−t .
This spectral analysis has a long history and many progresses have been made recently [Ra87, Dol98, Li04, Ts10, FaSj11, Ts12, BaLi12, GiLiPo13, DyZw13, FaTs13,
5In
this case, (Gt )t∈R is the Hamiltonian flow associated to p0 (x, ξ) :=
2
(|ξ�∗
x)
.
2
Lp NORMS, NODAL SETS, AND QUANTUM ERGODICITY
21
NoZw13]. We will not give any details on these advancements and we refer the interested reader to the above references.
In the present article, we will only make use of the results from [Li04], and for that
purpose, we recall a few definitions from this reference. We fix σ ∈ (0, γ0 ), and we
introduce the following dynamical distances between two points ρ, ρ� ∈ S ∗ M :
� +∞
s
�
d (ρ, ρ ) :=
eσt dS ∗ M (Gt (ρ), Gt (ρ� ))dt,
0
and
u
�
d (ρ, ρ ) :=
�
0
−∞
e−σt dS ∗ M (Gt (ρ), Gt (ρ� ))dt.
According to Lemma 2.3 in [Li04], du is a pseudo-distance on S ∗ M (meaning that it
can take the value +∞). Moreover, du , restricted to any strong-unstable manifold, is
a smooth function and it is equivalent to the restriction of the Riemannian metric,
while points belonging to diﬀerent unstable manifolds are at an infinite distance.
The analogous properties hold for ds . These distances allow to define diﬀerent
norms which are well adapted to the study of the transfer operator Lt . First, for
0 < β < 1, δ > 0 small enough and for every f in C 1 (S ∗ M ), we define
Hs,β (f ) :=
|f (ρ) − f (ρ� )|
,
ds (ρ, ρ� )β
ds (ρ,ρ� )≤δ
sup
and
|f |s,β := �f �C 0 + Hs,β (f ).
We observe that this norm is controlled by the standard β-H¨older norm, i.e. |f |s,β ≤
�f �C β . Then, we define Csβ (S ∗ M ) ⊂ C 0 (S ∗ M ) as the completion of C 1 (S ∗ M ) with
respect to the norm |.|s,β . We also introduce a family of test functions
Dβ := {f ∈ Csβ (S ∗ M ) : |f |s,β ≤ 1}.
Finally, we define the following norm, for every f in C 1 (S ∗ M ):
�f �β := �f �u + �f �s ,
with
��
�
�f �s := sup ��
g∈D
β
S∗M
�
�
f gdL�� , and �f �u := Hu,β (f ),
where L is the disintegration of the Liouville measure on S ∗ M , and Hu,β is defined
similarly as Hs,β except that we replace ds by du . We observe that
(22)
�f �s ≤ �f �L1 ≤ �f �C 0 .
In order to prove his main result on the exponential decay of correlations, Liverani
proved the following estimate (end of page 1284 in [Li04]):
�
�
(23)
∀t ≥ 0, �Lt (f )�s ≤ Cβ e−σβ t �X02 .f �β + �X0 .f �β + �f �β
`
HAMID HEZARI AND GABRIEL RIVIERE
22
�
for every β > 0 small enough, for every f ∈ C 3 (S ∗ M ) satisfying S ∗ M f dL = 0, and
for some positive constants Cβ > 0 and σβ > 0 (independent of f and t).
A.3. Multi-correlations. Thanks to these properties, we now prove the following
preliminary lemma:
Lemma A.1. Let p be a positive integer and let 0 < β < 1. Then, there exists
σp,β > 0 and Cp,β > 0 such that the following holds: �
For every τ > 0, for every f in C β (S ∗ M ) satisfying S ∗ M f dL = 0, and for every
0 ≤ t1 ≤ t2 ≤ . . . ≤ t2p satisfying
∃1 ≤ j0 ≤ 2p such that ∀k �= j0 , |tk − tj0 | ≥ τ,
one has
��
�
2p
�
�
�
�
�
tj
f ◦ G dL� ≤ Cp,β e−σp,β τ �f �2p
.
�
Cβ
� S∗M
�
j=1
Proof. We �start with the case of C 3 observables. Let f be an element in C 3 (S ∗ M )
such that S ∗ M f dL = 0, and let τ > 0. We fix6 1 ≤ j0 ≤ 2p − 1 such that
∀k �= j0 , |tk − tj0 | ≥ τ. We define
fj+0
:=
We set f˜j+0 := fj+0 −
�
and thus
�
2p
�
S ∗ M j=1
�
j0
�
j=1
f ◦G
tj −tj0
,
and, fj−0
j=j0 +1
f ◦ Gtj −tj0 +1 .
fj+0 dL. Then, we write
�
2p
�
tj
f ◦ G dL =
fj+0 ◦ Gtj0 −tj0 +1 fj−0 dL,
S∗M
S ∗ M j=1
tj
:=
2p
�
f ◦ G dL =
S∗M
�
S∗M
fj+0 dL
�
S∗M
fj−0 dL
+
�
S∗M
f˜j+0 ◦ Gtj0 −tj0 +1 fj−0 dL.
We now use the norms defined above to give an estimate on the remainder, i.e.
��
� �
� ��
�
�
� −�
+
−
t
−t
j
j
+1
˜+ �
˜ ◦ G 0 0 f dL� ≤ �
�
f
L
j0
� ∗ j0
� � tj0 +1 −tj0 fj0 �s Hs,β fj0 .
S M
Since tj − tj0 +1 ≥ 0 for every j ≥ j0 + 1, we get from the definition of Hs,β that
� �
0
Hs,β fj−0 = O(1)�f �2p−j
,
Cβ
where the constant does not depend on f and t = (t1 , . . . , t2p ). Similarly, as tj −tj0 ≤
0 for j ≤ j0 , using the definition of Hu,β and inequality (22), one can verify that,
6The
case j0 = 2p will be treated below.
Lp NORMS, NODAL SETS, AND QUANTUM ERGODICITY
23
for every l = 0, 1, 2, one has �X0l .f˜j+0 �β = O(1)�f �jC03 . Applying these observations
to Liverani’s estimate (23), we get that
�
2p
�
S ∗ M j=1
tj
f ◦ G dL =
�
S∗M
fj+0 dL
�
S∗M
fj−0 dL + O(1)e−σ0 (tj0 +1 −tj0 ) �f �2p
C3 ,
where the constant O(1) does not depend on f and t = (t1 , . . . , t2p ). If �j0 = 1, this
proves the result. In the case 2 ≤ j0 ≤ 2p, we have to analyze the term S ∗ M fj+0 dL.
For that purpose, we can write
�j −1
�
�
�
j0
0
�
�
tj −tj0
tj −tj0 −1
f ◦G
dL =
f ◦G
◦ Gtj0 −1 −tj0 f dL,
S ∗ M j=1
S∗M
j=1
and perform the same analysis to get the expected upper bound. Thus, we have
verified that there exists
Cp > 0 and σp > 0 such that, for every τ > 0, for every f in
�
3
∗
C (S M ) satisfying S ∗ M f dL = 0, and for every 0 ≤ t1 ≤ t2 ≤ . . . ≤ t2p satisfying
∃1 ≤ j0 ≤ 2p such that ∀k �= j0 , |tk − tj0 | ≥ τ,
one has
��
�
2p
�
�
�
�
�
2p
tj
f
◦
G
dL
�
� ≤ Cp e−σp τ �f �C 3 .
� S∗M
�
(24)
j=1
However recall that we are interested in getting a control in terms of H¨older norms.
To do this, we proceed
We fix 0 < β < 1 and a in
� as in Corollary 1 of σ[Dol98].
p
C β (S ∗ M ) such that S ∗ M adL = 0. Fix γ = 10p
. Using a convolution by a smooth
function, one can obtain a function f in C 3 (S ∗ M ) such that
�a − f �C 0 ≤ e−βγτ �a�C β , and �f �C 3 ≤ e3γτ �a�C β .
�
If we let f˜ = f − S ∗ M f dL, then it follows that there exists some constant C(p) > 0
such that
��
�
�
2p
2p
�
�
�
�
�
�
a ◦ Gtj dL −
f˜ ◦ Gtj dL� ≤ C(p)e−βγτ �a�2p
.
�
Cβ
� S∗M
�
S∗M
j=1
j=1
Applying (24), we find that there exists some constant C(p, β) > 0 such that
��
�
2p
�
�
�
�
�
a ◦ Gtj dL� ≤ C(p, β)(e(6pγ−σp )τ + e−βγτ )�a�2p
.
�
Cβ
� S∗M
�
j=1
Since γ =
σp
,
10p
the Lemma follows.
�
`
HAMID HEZARI AND GABRIEL RIVIERE
24
A.4. Proof of Proposition 2.3. Recall that we want to prove that, for every
0 < β < 1, there exist a constant Cβ,p > 0 such that, for every T > 0, and for every
a ∈ C β (S ∗ M ),
�� T
�2p
�
�
�
�
t
˜
�
� dL ≤ Cβ,p �a�2pβ T p .
Vp (a, T ) :=
a
◦
G
dt
−
adL
�
�
C
S∗M
S∗M
0
�
Without loss of generality, we can suppose that S ∗ M adL = 0 and that a is real
valued. We write
��
�
�
2p
�
V˜p (a, T ) =
a ◦ Gtj dL dt1 . . . dt2p .
[0,T ]2p
S ∗ M j=1
We now follow the idea of Lemma 3.2 of [Ra73], and define
�
�
A1 (T ) := (t1 , . . . , t2p ) ∈ [0, T ]2p : ∀1 ≤ j ≤ 2p, ∃k �= j : |tj − tk | < 1 .
For n ≥ 2, we also define
�
�
Bn (T ) := (t1 , . . . , t2p ) ∈ [0, T ]2p : ∀1 ≤ j ≤ 2p, ∃k �= j : |tj − tk | < n ,
and
We then clearly have
V˜p (a, T ) ≤
An (T ) = Bn (T ) − Bn−1 (T ).
��
n≥1
��
�
2p
�
�
�
�
�
a ◦ Gtj dL� dt1 . . . dt2p ,
�
�
An (T ) � S ∗ M
j=1
which by Lemma A.1 gives
�
�
�
(25)
V˜p (a, T ) ≤ Cp,β Leb(A1 (T )) +
Leb(An (T ))e−σp,β (n−1) �a�2pβ .
n≥2
C
It now remains to estimate Leb(An (T )) for every n ≥ 1. For every n ≥ 1, using a
simple rescaling we get
(26)
Leb(An (T )) ≤ Leb(Bn (T )) = n2p Leb(A1 (T /n)) ≤ n2p Leb(A1 (T )).
Hence we only have to estimate Leb(A1 (T )). To do this we write
�
�
�
2p
�
�
Leb(A1 (T )) ≤
1[−1,1] (tj − tk ) dt1 . . . dt2p .
�2p
��
[0,T ]2d j=1
k�=j
�
The function j=1
k�=j 1[−1,1] (tj − tk ) does not vanish if and only if we can split
the 2p-uple (t1 , . . . , t2p ) into N disjoint subfamilies (tα11 , . . . , tα1n(1) ), . . . , (tαN1 , . . . , tαNn(N ) )
such that
• for every 1 ≤ j ≤ N , n(j) ≥ 2,
Lp NORMS, NODAL SETS, AND QUANTUM ERGODICITY
25
• for every tαjr element in the j-th subfamily, there exists 1 ≤ r� �= r ≤ n(j)
such that |tαjr − tαj � | ≤ 1.
r
It implies that there exists some universal constant cp > 0 (depending only on p)
such that Leb(A1 (T )) ≤ cp T p . This, together with (25) and (26), implies that
V˜p (a, T ) ≤ C˜p T p �a�2p
.
Cβ
Acknowledgments
The first author is partially supported by the NSF grant DMS-0969745. The second
author is partially supported by the Agence Nationale de la Recherche through the
Labex CEMPI (ANR-11-LABX-0007-01) and the ANR project GeRaSic (ANR-13BS01-0007-01).
References
[AnRi12] N. Anantharamn, G. Rivi`ere Dispersion and controllability for the Schr¨
odinger equation
on negatively curved manifolds. Analysis and PDE 5 (2012), 313–338.
[BaLi12] V. Baladi, C. Liverani Exponential decay of correlations for piecewise cone hyperbolic
contact flows. Comm. Math. Phys. 314 (2012), no. 3, 689–773.
[B´e77] P. H. B´erard On the wave equation on a compact Riemannian manifold without conjugate
points. Math. Z. 155 (1977), no. 3, 249–276.
[BlSo13] M. Blair and C. Sogge On Kakeya-Nikodym averages, Lp -norms and lower bounds for
nodal sets of eigenfunctions in higher dimensions, to appear in Journal of the European
Mathematical Society. 2013.
[Bo93] J. Bourgain Eigenfunctions bounds for the Laplacian on the n-torus. IMRN 3 (1993),
61–66.
[Bo13] J. Bourgain Moment inequalities for trigonometric polynomials with spectrum in curved
hypersurfaces, Israel J. Math. 193 (2013), 441–458
[BoDe14] J. Bourgain and C. Demeter The proof of the l2 decoupling conjecture. arXiv:1403.5335
(2014)
¨
[Br78] J. Br¨
uning Uber
Knoten Eigenfunktionen des Laplace-Beltrami Operators. Math. Z. 158
(1978), 15–21.
[CoMi11] T. Colding and W. P. Minicozzi II Lower bounds for nodal sets of eigenfunctions. Comm.
Math. Phys. 306 (2011), no. 3, 777–784.
[CdV85] Y. Colin de Verdi`ere Ergodicit´e et fonctions propres du Laplacien, Comm. in Math. Phys.
102, 497–502 (1985)
[Dol98] D. Dolgopyat On decay of correlations in Anosov flows. Ann. of Math. (2) 147 (1998),
no. 2, 357–390.
[D92]
R. T. Dong Nodal sets of eigenfunctions on Riemann surfaces. J. Diﬀerential Geom. 36
(1992), no. 2, 493–506.
[DoFe88] H. Donnelly and C. Feﬀerman Nodal sets of eigenfunctions on Riemannian manifolds.
Invent. Math. 93 (1988), 161–183.
26
`
HAMID HEZARI AND GABRIEL RIVIERE
[DoFe90] H. Donnelly and C. Feﬀerman Nodal sets for eigenfunctions of the Laplacian on surfaces.
J. Amer. Math. Soc. 3 (1990), 332–353.
[DuGu75] J.J. Duistermaat and V. Guillemin, The spectrum of elliptic operators and periodic
bicharacteristics, Inv. Math. 29 (1975), 39–79
[DyZw13] S. Dyatlov, M. Zworski Dynamical zeta function for Anosov flows via microlocal analysis,
preprint arXiv:1306.4203 (2013)
[FaSj11] F. Faure, J. Sj¨
ostrand Upper bound on the density of Ruelle resonances for Anosov flows,
Comm. in Math. Physics. 308 (2011), 325–364.
[FaTs13] F. Faure, M. Tsujii The semiclassical zeta function for geodesic flows on negatively curved
manifolds, preprint arXiv:1311.4932 (2013).
[GiLiPo13] P. Giulietti, C. Liverani, M Pollicott Anosov Flows and Dynamical Zeta Functions,
Annals of Mathematics 178, 2 (2013), 687–773.
[Han14] X. Han Small scale quantum ergodicity on negatively curved manifolds. arXiv:1410.3911
(2014).
[HaTa13] A. Hassell and M. Tacy Improvement of eigenfunction estimates on manifolds of nonpositive curvature. arXiv:1212.2540, to appear in Forum Mathematicum.
[HeMaRo87] B. Helﬀer, A. Martinez, D. Robert Ergodicit´e et limite semi-classique, Comm. in
Math. Phys. 109 (1987), 313–326.
[HeRi15] H. Hezari and G. Rivi`ere Quantitative equidistribution properties of toral eigenfunctions.
preprint arXiv:1503.02794 (2015)
[HeSo12] H. Hezari and C. D. Sogge A natural lower bound for the size of nodal sets. Anal. PDE
5 (2012), no. 5, 1133–1137.
[IwSa95] H. Iwaniec, P. Sarnak L∞ norms of eigenfunctions of arithmetic surfaces. Ann. Math.
141 (1995), 301–320.
[Ju13] J. Jung Quantitative quantum ergodicity conjecture and the nodal domains of MaassHecke cusp forms. Preprint Arxiv 1301.6211 (2013).
[KoTaZw07] H. Koch, D. Tataru, and M. Zworski Semiclassical Lp estimates . Ann. Henri Poincar´e
8 (2007), no. 5, 885–916.
[Li04]
C. Liverani On contact Anosov flows. Ann. Math. 159 (2004), 1275–1312.
[LuSa95] W. Luo, P. Sarnak Quantum ergodicity of eigenfunctions on P SL2 (Z)\H2 . Publ. Math.
´ 81 (1995), 207–237.
IHES
[MaRu12] J. Marklof and Z. Rudnick Almost all eigenfunctions of a rational polygon are uniformly
distributed. J. of Spectral Theory. 2 (2012), 107–113.
[MeTo12] I. Melbourne, A. T¨
or¨
ok Convergence of moments for axiom A and non-uniformly hyperbolic flows, Ergodic Theory Dynam. Systems 32 (2012), no. 3, 1091–1100.
[NoZw13] S. Nonnenmacher, M. Zworski Decay of correlations for normally hyperbolic trapping,
to appear in Inv. Mathematicae.
[Ra73] M. Ratner The central limit theorem for geodesic flows on n-dimensional manifolds, Israel
J. Math. 16 (1973), 181–197.
[Ra87] M. Ratner The rate of mixing for geodesic and horocycle flows, Ergodic Theory Dynam.
Systems 7 (1987), 267–288.
[Ri13] G. Rivi`ere Remarks on quantum ergodicity, Journal of Modern Dynamics, 7 (2013), 119–
133.
[Sch06] R. Schubert, Upper bounds on the rate of quantum ergodicity Ann. Henri Poincar´e 7
(2006), 1085–1098.
[Sh74] A. Shnirelman Ergodic properties of eigenfunctions, Usp. Math. Nauk. 29 (1974), 181–
182.
[SmZw13] H. Smith and M. Zworski Pointwise bounds on quasimodes of semiclassical Schr¨
odinger
operators in dimension two. Math. Res. Lett. 20 (2013), 401–408.
Lp NORMS, NODAL SETS, AND QUANTUM ERGODICITY
27
C. Sogge Concerning the Lp norm of spectral clusters for second-order elliptic operators
on compact manifolds. J. Funct. Anal., 77 (1):123–138, 1988.
[So93] C. D. Sogge Fourier integrals in classical analysis. Cambridge Tracts in Mathematics,
105. Cambridge University Press, Cambridge, 1993.
[So11] C. D. Sogge Kakeya-Nikodym averages and Lp-norms of eigenfunctions. Tohoku Math.
J. (2), 63 (4):519–538, 2011.
[So15] C. D. Sogge Localized Lp -estimates of eigenfunctions: A note on an article of Hezari and
Rivi`ere. preprint arXiv:1503.07238, 2015.
[SoToZe11] C. D. Sogge, J. A. Toth, and S. Zelditch About the blowup of quasimodes on Riemannian manifolds. J. Geom. Anal., 21(1):150–173, 2011.
[SoZe11] C. Sogge and S. Zelditch On eigenfunction restriction estimates and L4 -bounds for compact surfaces with nonpositive curvature. arXiv:1108.2726v2, 2011.
[SoZe11] C. D. Sogge and S. Zelditch Lower bounds on the Hausdorﬀ measure of nodal sets. Math.
Res. Lett. 18 (2011), no 1, 25–37.
[SoZe12] C. D. Sogge and S. Zelditch Lower bounds on the Hausdorﬀ measure of nodal sets II.
Math. Res. Lett. 19 (2012), no. 6, 1361–1364.
[SoZe12b] C. D. Sogge and S. Zelditch Concerning the L4 norms of typical eigenfunctions on
compact surfaces. Recent developments in geometry and analysis, 407–423, Adv. Lect.
Math. 23, Int. Press, Somerville, MA, (2012).
[Ts10] M. Tsujii Quasi-compactness of transfer operators for contact Anosov flows. Nonlinearity
23 (2010), no. 7, 1495–1545.
[Ts12] M. Tsujii Contact Anosov flows and the Fourier-Bros-Iagolnitzer transform. Ergodic Theory Dynam. Systems 32 (2012), no. 6, 2083–2118.
[Ya82] S.-T. Yau Survey on partial diﬀerential equations in diﬀerential geometry. Seminar on
Diﬀerential Geometry, 3–71, Ann. of Math. Stud., 102, Princeton Univ. Press, Princeton,
N.J., 1982.
[Yo13] M.-P. Young The quantum unique ergodicity conjecture for thin sets. Preprint Arxiv
1306.1554 (2013).
[Ze87] S. Zelditch Uniform distribution of the eigenfunctions on compact hyperbolic surfaces,
Duke Math. Jour. 55, 919–941 (1987)
[Ze94] S. Zelditch On the rate of quantum ergodicity. I. Upper bounds. Comm. Math. Phys. 160
(1994), no. 1, 81–92.
[Ze96] S. Zelditch Quantum ergodicity of C ∗ dynamical systems. Comm. Math. Phys. 177 (1996),
no. 2, 507–528.
[Ze08] S. Zelditch Local and global analysis of eigenfunctions. A survey on eigenfunctions of
the Laplacian on Riemannian manifolds. Advanced Lectures in Mathematics (ALM) 7,
545–658 (2008), arXiv:0903.3420v1.
[Ze13] S. Zelditch Eigenfunctions and nodal sets. Geometry and topology, 237–308, Surv. Diﬀer.
Geom., 18, Int. Press, Somerville, MA, 2013.
[Zw12] M. Zworski Semiclassical analysis. Graduate Studies in Mathematics 138, AMS (2012).
[Zy74] A. Zygmund, On Fourier coeﬃcients and transforms of functions of two variables. Studia
Mathematica, 50 (1974), no 2, 189–201.
[So88]
28
`
HAMID HEZARI AND GABRIEL RIVIERE
Department of Mathematics, UC Irvine, Irvine, CA 92617, USA
E-mail address: [email protected]
´ (U.M.R. CNRS 8524), U.F.R. de Mathe
´matiques, UniverLaboratoire Paul Painleve
´
site Lille 1, 59655 Villeneuve d’Ascq Cedex, France
E-mail address: [email protected]